Jed Brown

The updated libffi is needed to fix an issue with MacOS 15, and the fix has a side-effect of increasing MSRV to 1.78. See https://github.com/rsmpi/rsmpi/pull/209

offline post-processing of collapsed stacks without running a workload

1

I often need to analyze collapsed stacks produced using different tools or on different systems. This can be done today with a slightly awkward command (cf. https://github.com/flamegraph-rs/flamegraph/issues/259#issuecomment-1579859541) ```console $ flamegraph...

rust - re-export libceed-sys features through libceed to allow building non-static without using libceed-sys directly

11

Fix #1694 I will note that there is nothing wrong with depending on `libceed-sys` directly to set the feature at that level, so I don't think this is a necessary...

Lightweight local job submission (~ 10ms)

22

The Python-based `flux mini run` costs over 100 ms, and thus becomes a bottleneck in some scenarios, including unit testing with independent executable invocations. This was discussed off-topic in https://github.com/flux-framework/flux-core/issues/2915...

add nvptx_target_feature and use unaligned barrier

4

* Deprecate `_syncthreads` (the CUDA name) in favor of new `_barrier_sync` (NVPTX name `barrier.sync`). * The: `barrier.sync` instruction [is equivalent](https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-bar) to `barrier.sync.aligned` prior to `sm_70`, and will lead to errors/deadlock...